kami
118ac9fbcb
Not my work. These nine files sat uncommitted in the shared checkout while another session worked on them, and the UI redesign that follows rewrites web/src/main.tsx and web/src/style.css. Committing first means that work is recoverable rather than overwritten. Contents, by inspection rather than by authorship: whitespace normalisation and edits across main.tsx, 568 added lines of style.css, client and client test changes, the orchestra-user line in build.sh, and docs updates to AGENTS.md, AUDIT.md, DEPLOYMENT.md and the env example. Committed at the operator's explicit instruction.
4.1 KiB
Deployment verification
Both binaries embed their Git revision, UTC build time, and dirty flag. The
coordinator is deployed as a Docker Compose image — see "For the Docker
coordinator deployment" below for the build that carries provenance. (The old
deploy/redeploy.sh + orchestra.service path was deleted on 2026-07-31;
orchestra-worker.service is a different, still-current unit.)
Browser operator login
Browser operators now live in the embedded ${ORCHESTRA_DATA}/auth.db
database. Passwords are bcrypt-hashed inside that database; no password hash
belongs in .env.
For a new local data directory, create the first account while Orchestra is stopped. The command reads and confirms the password from the terminal:
go run ./cmd/orchestra-user set -data ./data -username kami
For the Docker Compose deployment, the API image includes the same helper. Keep the API stopped while it opens the database, then use the existing data volume through Compose:
docker compose stop orchestra-api
docker compose run --rm --entrypoint /app/orchestra-user \
orchestra-api set -data /data -username kami
docker compose up -d orchestra-api
After signing in, the Settings screen can change the username or password.
Every browser session for that account is revoked after a credential change.
To recover a forgotten password, stop the API and run orchestra-user set
again for the same username. orchestra-user list -data /data lists usernames
without exposing password hashes.
On the first start after upgrading, an empty auth database automatically
imports the existing ORCHESTRA_WEB_USERNAME and
ORCHESTRA_WEB_PASSWORD_HASH pair. Once the startup log confirms the import,
remove both legacy values from .env; they are ignored whenever the database
already contains an account. ORCHESTRA_WEB_TOKEN remains unused by the
browser UI.
Build a worker for staging on workpc with:
revision=$(git rev-parse HEAD)
build_time=$(date -u +%Y-%m-%dT%H:%M:%SZ)
dirty=false; test -z "$(git status --porcelain)" || dirty=true
go build -ldflags "-X orchestra/internal/buildinfo.Revision=$revision -X orchestra/internal/buildinfo.Time=$build_time -X orchestra/internal/buildinfo.Dirty=$dirty" -o orchestra-worker ./cmd/orchestra-worker
scp orchestra-worker workpc:~/orchestra-deploy/orchestra-worker
ssh workpc 'sha256sum ~/orchestra-deploy/orchestra-worker'
The worker receives only the path to a normal project configuration file:
ORCHESTRA_WORKER_PROJECT_CONFIG_FILE=/etc/orchestra/worker-projects.json.
That file contains a JSON object whose project entries contain repo,
worktree_root, and remote; mount or provision it like any other worker
configuration. The legacy single-checkout ORCHESTRA_WORKER_PROJECTS comma
list remains supported for one existing checkout. An absent project is
ineligible for routing.
Verify the coordinator at GET /v1/admin/diagnostics with the normal admin
credential: its build object is the coordinator provenance. GET /v1/federation/workers shows every worker's build, supported projects, and
worker-local health without SSH.
Build the coordinator and worker with deploy/build.sh, which stamps them from one commit
and refuses a dirty tree. A burn-in run must never pair a new coordinator with
an old worker, and matching revisions are how that is checked rather than
assumed.
For the Docker coordinator deployment, provide the same provenance as build
arguments (the Dockerfile intentionally cannot read .git from its build
context):
revision=$(git rev-parse HEAD)
build_time=$(date -u +%Y-%m-%dT%H:%M:%SZ)
dirty=false; test -z "$(git status --porcelain)" || dirty=true
docker compose build \
--build-arg BUILD_REVISION="$revision" \
--build-arg BUILD_TIME="$build_time" \
--build-arg BUILD_DIRTY="$dirty" \
orchestra-api
docker compose up -d --no-deps orchestra-api
If a pre-v2 event log has the historical repeated-seq=1 prefix, the current
coordinator intentionally refuses to replay it. Stop every coordinator using
the data directory and run the explicit, backup-preserving migration before
deploying the current image:
orchestra-migrate -data /var/lib/orchestra/data -confirm